Identity agent: Store abstraction with in-memory fallback

[bokelley]

Summary

• Extract Store interface abstracting sorted-set, set, and KV operations used by the identity agent
• RedisStore wraps go-redis with StorePipeline for batched commands
• InMemoryStore provides graceful fallback when Valkey is unavailable
• Auto-detect Valkey connectivity on startup, fall back to in-memory
• Health endpoint (GET /health) reports store status
• Fix all pre-existing lint issues

Closes #5

Test plan

• All existing tests pass (using RedisStore + miniredis)
• New in-memory store tests: full frequency cap flow, audience segments
• Tests pass with -race
• golangci-lint clean

🤖 Generated with Claude Code

[bokelley]

Leaving this open intentionally as the architectural reference for the exposure-storage design discussion happening on #68.

The Store interface here (ZAdd/ZCount/Expire + Pipeline + in-memory fallback) is exactly the shape #68 is arguing against in the production targeting/ package. Before rebasing the layout changes (reference/identity-agent was restructured into cmd/identity-agent/ after this was filed), we need to resolve: does the production Store keep sorted-set primitives, or not?

• If yes → this PR's design is correct, rebase it onto the new layout and merge.
• If no → close this and document the new exposure-storage design before refactor(targeting): remove RecordExposure path #68 lands.

Tagging for resolution alongside #68.

[bokelley]

Closing — premise is superseded.

PR #75 (merged 2026-04-22) removed the sorted-set primitives (ZAdd / ZCount / Expire / ZRemRangeByScore) from the production Store interface, which was the abstraction this PR built on. The reference layout also moved from reference/identity-agent/ to cmd/identity-agent/, so this branch has been CONFLICTING against a moving target for a month.

The in-memory fallback idea is still useful but belongs in a fresh, small PR against the current targeting/store.go shape — not a rebase of this 395-line diff. Will file a separate issue if/when we want it.

[bokelley]

Acknowledged — closing rationale noted. The in-memory fallback concept is tracked for a future standalone issue against the current targeting/store.go shape whenever you're ready to pursue it.

Triaged by Claude Code. Session: https://claude.ai/code/${CLAUDE_CODE_REMOTE_SESSION_ID}

---

Generated by Claude Code